Re: Re: [mpls-linux-general] binding label/FEC to hosts

[krishnan. N <kri...@my...>]

Hello,

> 
> so you have one ingress label.
>
   Maybe, maybe more.
 
> >     I am curious to know:
> >     a) Would it be useful to be able to bind a label to
a host given
> >        only a net entry in the routing table?
> 
> Hmm, why add another layer above ARP?
> 
>  You only have one ingress label -> you *must* look at the 
> ip header to resolve the L2 destination, you cant infer
> it from the label.
> 
> From this ip header, you want another token (a label). 
> There is already a similar protocol, which resolves an 
> ip address to a flat token, it is called ARP.
> 
   There appears to be some confusion i have created; let
   me try again.

   The routing table has entries aggregated, i.e, all the
   routes a sub-net are resolved into a single entry
   
   192.168.2.0  gw 192.168.1.2 

   So, we need to go thru host with interface 1.2 to
   reach all hosts in the 2.0 interface. Let us assume 
   that there are 3 hosts 2.1, 2.2 and 2.3 in that sub-net.

   What i wanted to do was be able to say that only 
   packets to, say, 192.168.2.3 should be MPLS switched.

   In the current implementation, it appears that i can
   only specify an FEC with an existing entry in the
   routing table. I can only set up an FEC mapping:
 
mplsadm -A -B -O gen:16:eth0:192.168.2.1 -f 192.168.2.0/24
                                                    ^^^^^^
  and not:

mplsadm -A -B -O gen:16:eth0:192.168.2.1 -f 192.168.2.2/32 
                                                    ^^^^^^

  In any normal setup i would imagine that, when routing
  protocols are running, all routes will be aggregated 
  and MPLS'ing a host a few hops away would be required.

  Hope this makes my question a little more comprehensible.

thank you,
shyam